Thoughts and Assumptions about the given data:
The given data has app, bid_price ,win, events as the columns. The bid_price is a continuous column in nature. In practical scenario, we can bid with any continuous value > 0.
In practical scenario, we can bid for any price but in the given table we have price in [0.01, 0.1, 0.2, 0.4, 0.5, 0.75, 1, 2, 5, 9] .
Naturally,
- bid_price > 0 ( bid_price will always be greater 0 otherwise it does not makes sense )
- bid_price < 9 ( since we already got win_rate of 100% with this bid_price)
Note : 100% win_rate would mean at this bid, there is a 100% chance we will win.
Observations (for app A):
- There is a 10x change in number of events with every every bid_price.
- The most frequent bid_prices are ~ 0.2.
Assumption:
- Maximum number of bids are at 0.2 bid_price i.e. we are considering 0.2 has the value with maximum bids that were made
Here is the table which has the win_rate for the price mentioned :
| app | bid_price | winrate | total_events |
|---|---|---|---|
| A | 0.01 | 0 | 100000 |
| A | 0.1 | 0.3 | 10000 |
| A | 0.2 | 0.2 | 10000000 |
| A | 0.4 | 0.3 | 1000000 |
| A | 0.5 | 0.2 | 100000 |
| A | 0.75 | 0.3 | 10000 |
| A | 1 | 0.6 | 1000 |
| A | 2 | 0.7 | 100 |
| A | 5 | 0.8 | 10 |
| A | 9 | 1 | 1 |
The size of the dots denote the total events
The simple answer is to maximize the net_revenue , we have to minimize the bid_price.
In the original table and considering the win_rate. 0.2 bid_price has just win_rate of 20%, but the total_events = 10,000,000. So basically, we won 2,000,000 times.
The most optimal bid_valuation that we should send ~ 0.2 because even if the advertiser is willing to pay 0.21$ we get net_revenue as 0.01 * 2000000 = 20,000$ .
Having said that, I would also investigate more around 0.1$ bid_price by doing more bids around the value to see how the winrate changes if we increase the number of total_events.